CMRT 26: Worker: Port FingerprintMatcher from PR #15585 POC

[Swarnadip-Kar]

Turns Daniel's matchFingerprints() POC (PR mixxxdj#15585) into a real,
non-debug-gated class: FingerprintMatcher. This is the comparison engine
CMRT 27's grouping pipeline calls -- nothing wires it up yet, this PR is
just the algorithm itself plus its own constants.

Public API

• simHashCandidatesMatch(hashA, hashB, maxHammingBits) -- O(1) pre-filter
• compare(a, b) -> MatchResult{isMatch, score, offsetItems}
• kItemDurationSeconds (~0.1238s/item) and kSimHashMaxHammingBits (8) as
  public constants, since CmrtGroupingService needs the same values

Algorithm, split into two phases (private)

• findTopOffset() -- bucket-vote coarse alignment search (POC's first loop)
• scoreFingerprints() -- calls findTopOffset(), then fine-offset refinement
  (the sin() correction term), the 2%-of-unique-size uniqueness guard, and
  the XOR/popcount bit-error score with diversity correction for
  repetitive audio

Ported as-is, not fixed (commented, not silently changed)

• UNIQ_BITS/UNIQ_MASK/UNIQ_STRIP in the POC reference MATCH_BITS instead of
  the UNIQ_BITS they declare -- kept the same bucket width rather than
  changing what "duplicate item" means without a deliberate decision to.
• The bit-error pass's reinterpret_cast to quint64* assumes 8-byte
  alignment, same as the POC. Works in practice; flagged with TODO(XXX)
  rather than adding an aligned-copy step this PR doesn't otherwise need.

Fixed, not ported as-is

• The POC's memset(seen, 0, UNIQ_MASK) only clears half of the backing
  array (wrong byte count for a uint16_t[]). QVector::fill(0) here clears
  correctly.

Notes for reviewers

• kMaxOffsetSearchItems = 120 is the value the POC's own test calls used.
  No other guidance exists yet for what the production search window
  should be -- flagging this as the one constant most likely to need
  re-tuning once this runs against a real library.
• findTopOffset() is exposed as a separate private method (not just inlined
  into scoreFingerprints()) so it can get its own unit test later without
  needing fixture fingerprints large enough to exercise the full scoring
  pass.

Depends on: cmrt-quality-scoring (#34)

[coveralls]

Coverage Report for CI Build 27872408352

Coverage remained the same at 30.438%

Details

• Coverage remained the same as the base build.
• Patch coverage: No coverable lines changed in this PR.
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	116230
Covered Lines:	35378
Line Coverage:	30.44%
Coverage Strength:	59300.22 hits per line

---

💛 - Coveralls